Forecasting spot electricity prices by deep learning techniques: An application on Turkish Energy Exchange

Turkish electricity market has transformed from a vertically integrated structure to a competitive market with reforms and deregulations. As a result of this, in March 2015, Energy Markets Operations Inc. (EPİAŞ) was established. Within EPİAŞ, Day-Ahead Market and Intra-Day Market, where physical electricity trade is made, are operated. The Day-Ahead Market and the Intraday Market are the spot markets of electricity. The Day Ahead Market offers trading opportunities for market participants in addition to bilateral agreements and makes a significant contribution to the market balance one day ahead of the real time. The Market Clearing Price formed in the Day Ahead Market is defined as the reference price of electrical energy. In this sense, for each hour a Market Clearing Price is determined in the Day Ahead Market. Forecasting the hourly Market Clearing Price is important for market participants to develop bid strategies and to plan production/supply programs. In this study, it is aimed to forecast hourly Market Clearing Price realized in the Day Ahead Market, which is operated by Turkish Energy Exchange EPİAŞ. In this direction, forecastings for 24 hours and 168 hours ahead are made by Multilayer Perceptrons (MLP), Convolutional Neural Networks (CNN), Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) models, The findings show that for 24 hours ahead forecasting MLP, CNN, LSTM and GRU models create 8,44, 9,27, 8,15 and 8,72 average MAPE values, respectively. And, for the 168 hours ahead forecasting, the average MAPE values obtained by MLP, CNN and GRU models are 10,60, 10,13 and 9,58, respectively. Key Words: Electricity Price Forecast, Day Ahead Market, Market Clearing Price, Deep Learning, EPİAŞ.